AI summary · may contain errors
Biomni, a general-purpose biomedical AI agent, autonomously executes diverse research tasks and shows strong generalization across heterogeneous domains without task-specific tuning.
- 4 cited this week
Moving in Science.
Rebuilt
AI summary · may contain errors
Biomni, a general-purpose biomedical AI agent, autonomously executes diverse research tasks and shows strong generalization across heterogeneous domains without task-specific tuning.
Newest first · last 60 days
AI summary · may contain errors
EcoXAI, a containerized multi-agent framework for biomedical discovery, evaluated 103 Alzheimer’s Disease drug candidates and identified 79 novel candidates with predictive models exceeding a randomized baseline.
AI summary · may contain errors
Agent3Diso is a hybrid deterministic–agent multimodal framework that resolves disease-relevant molecular forms across transcript-to-structure layers, enabling modality-specific therapeutic target discovery.
AI summary · may contain errors
RESCUE, a multi-agent LLM workflow in a pediatric hospital EHR, achieved AUC 0.808 on a holdout set of n=12,591 and flagged higher genomic-assessment follow-up.
AI summary · may contain errors
Ontology-based integration of 9,147 vaccine trials into SEA-CDM linked to Vaccine Ontology and OAE enables semantic retrieval and improved adverse-event ranking, including recall@10 rising from 0.317 to 0.383.
AI summary · may contain errors
SLiMMine, a deep learning method for the human proteome, identifies functional short linear motifs while eliminating ~80% of pattern-matching false-positive hits.
AI summary · may contain errors
PHENOCAUZ maps 7,152 human symptoms to molecular determinants across 4,222 Mendelian proteins, achieving ~0.60 precision among the top 20 predictions for well-characterized symptoms.
AI summary · may contain errors
An iterative, cross-species harmonization effort produced HOMBA, a formal ontology and 3D-mappable basal ganglia framework spanning human, macaque, and marmoset brains.
AI summary · may contain errors
PGCopilot, an AI-assisted research agent for plant functional genomics, integrates planning and tool orchestration to retrieve and synthesize evidence from 1.83 million records and 33 million PubMed-indexed articles.
AI summary · may contain errors
CZ CELLxGENE Discover scaled into a widely used single-cell genomics community resource by pairing contributors with dedicated curators to improve metadata quality and richness.
AI summary · may contain errors
ProtoPure, an LLM-based literature-mining workflow, extracts standardized protein purification conditions and finds convergence in recombinant expression practices, with BL21(DE3) systems dominating across thousands of proteins.
AI summary · may contain errors
Claude Opus 4.6 most accurately extracted glaucoma prior and future procedure data from free-text EHR notes, reaching 0.996 accuracy for prior procedures and 0.978 for future procedures.
AI summary · may contain errors
GLESKit, an MCP-based gene-centered system, completed all 15 GeneReportBench reports with a mean committee score of 0.904 and answered 35 of 39 BixBench questions.
AI summary · may contain errors
A temporally resolved interactomic atlas in a mouse tauopathy model maps phosphorylation-dependent remodeling of multi-protein complexes across disease progression, capturing composition, dynamics, and regulatory state at scale.
AI summary · may contain errors
GPT 4.1-mini achieved the best overall extraction performance on 200 synthetic mock molecular test reports, reaching an average F1 score of 79.2% across quality levels and modalities.
AI summary · may contain errors
BRIDGE prioritizes human–mouse gene correspondences, achieving 61.8–66.7% Recall@5 across three functional-correspondence benchmarks using disease and functional evidence beyond orthology alone.
AI summary · may contain errors
CTD now contains over 110 million integrated toxicogenomic connections linking harmonized chemicals, genes/proteins, phenotypes, diseases, pathways, and human exposures.
AI summary · may contain errors
An automated literature-mining workflow built a CAR T cell signaling knowledge graph with ~7,500 unique interactions among ~1,800 entities from PubMed.
AI summary · may contain errors
An ensemble NLP model using Bio_ClinicalBERT on radiology report Findings and Impressions detects brain metastases with sensitivity 0.888 at precision 0.499 in internal Alberta validation.
AI summary · may contain errors
A privacy-preserving Meta Llama 3.3 70B pipeline extracted over 50 oncology registry variables from 6,700 Singapore cancer-center notes with high F1 scores, peaking at diagnosis (97.2%).
AI summary · may contain errors
Frontier large language models can quantitatively summarize literature-wide neuroscience, revealing structure–function mappings across brain regions that existing meta-analytic frameworks miss, including via lesions and intracranial stimulation.
AI summary · may contain errors
LitGene, an interpretable transformer integrating text and Gene Ontology, achieves competitive accuracy across eight protein-property benchmarks and enables zero-shot prediction of GO terms, KEGG pathways, and gene-disease associations.
AI summary · may contain errors
CPIExtract, an open-source Python package for compound–protein interaction extraction, aggregates experimental binding annotations and collects more than ten times the number versus DrugBank alone.
AI summary · may contain errors
In TCGA RNA-seq tumour samples, GO-constrained biologically-informed neural networks add little to reconstruction or latent-space organization, with performance often matched by randomized or unconstrained models.
AI summary · may contain errors
PGRanker ranks 21,592 candidate genes without variant data by seeding graph learning from direct HPO gene support, improving mean reciprocal rank to 0.192 on 277 Mendelian cases.
AI summary · may contain errors
AXIS reconstructs published QSP simulation scenarios from a companion paper and model, recovering reported scenarios and producing parameter assignments matching expert-curated ground truth.
AI summary · may contain errors
ddkg.skill enables correct Cypher querying of the NIH Common Fund Data Ecosystem Data Distillery Knowledge Graph, achieving correct executed results in 5 of 7 tests for absent sources.
AI summary · may contain errors
MOSAIC, an unsupervised graph embedding method, achieved node-type classification AUROCs of 0.9863–0.9926 on a biomedical graph with 5,171 human genes and 10,147 phenotypes.
AI summary · may contain errors
SNACKKSS variant SA4 uniquely improves predictions of protein-inhibiting compounds via regulatory relationship modeling, adding value beyond existing predictors in cross-validation.
Journals this month
Topics moving, week to 18 Nov 2026
By field